Blood pressure and risk of cancer: a Mendelian randomization study

(2021) 21:1338

Chan et al. BMC Cancer



Open Access

RESEARCH

Blood pressure and risk of cancer:

a Mendelian randomization study

Io Ieong Chan1, Man Ki Kwok1 and C. Mary Schooling1,2*

Abstract

Background: Previous large observational cohort studies showed higher blood pressure (BP) positively associated

with cancer. We used Mendelian randomization (MR) to obtain less confounded estimates of BP on total and sitespecific cancers.

Methods: We applied replicated genetic instruments for systolic and diastolic BP to summary genetic associations

with total cancer (37387 cases, 367856 non-cases) from the UK Biobank, and 17 site-specific cancers (663¨C17881

cases) from a meta-analysis of the UK Biobank and the Kaiser Permanente Genetic Epidemiology Research on Adult

Health and Aging. We used inverse-variance weighting with multiplicative random effects as the main analysis, and

sensitivity analyses including the weighted median, MR-Egger and multivariable MR adjusted for body mass index

and for smoking. For validation, we included breast (Breast Cancer Association Consortium: 133384 cases, 113789

non-cases), prostate (Prostate Cancer Association Group to Investigate Cancer Associated Alterations in the Genome

Consortium: 79194 cases, 61112 non-cases) and lung (International Lung and Cancer Consortium: 10246 cases, 38295

non-cases) cancer from large consortia. We used asthma as a negative control outcome.

Results: Systolic and diastolic BP were unrelated to total cancer (OR 0.98 per standard deviation higher [95% confidence interval (CI) 0.89, 1.07] and OR 1.00 [95% CI 0.92, 1.08]) and to site-specific cancers after accounting for multiple

testing, with consistent findings from consortia. BP was nominally associated with melanoma and possibly kidney

cancer, and as expected, not associated with asthma. Sensitivity analyses using other MR methods gave similar results.

Conclusions: In contrast to previous observational evidence, BP does not appear to be a risk factor for cancer,

although an effect on melanoma and kidney cancer cannot be excluded. Other targets for cancer prevention might

be more relevant.

Keywords: Blood pressure, Cancer, Mendelian Randomization

Background

Elevated blood pressure (BP), or hypertension, reduces

population health globally [1], with 31.1% of the world¡¯s

adult population estimated to be hypertensive in 2010

[2], and 10.4 million deaths worldwide attributed to

high systolic BP in 2016 [3]. In addition to the wellestablished relation of BP with cardiovascular disease

*Correspondence: cms1@hku.hk

1

School of Public Health, Li Ka Shing Faculty of Medicine, The University

of Hong Kong, 7 Sassoon Road, Pokfulam, Hong Kong, SAR, China

Full list of author information is available at the end of the article

(CVD) [4], hypertension has been linked with higher

risk of cancer observationally [5¨C7], but the evidence

is inconsistent with the possible exception of kidney

cancer [8]. Secondary analyses of randomized controlled trials (RCTs) of antihypertensive drugs found

little association with cancer [9], but RCTs typically

have follow-up times too short to detect effects on cancer risk. Although the underlying mechanisms linking

hypertension to cancer are still unclear, it has been suggested that increased cell turnover and telomere shortening could play a role [10]. In addition, dysregulated

immune function is implicated in the pathogenesis

? The Author(s) 2021. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which

permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the

original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or

other third party material in this article are included in the article¡¯s Creative Commons licence, unless indicated otherwise in a credit line

to the material. If material is not included in the article¡¯s Creative Commons licence and your intended use is not permitted by statutory

regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this

licence, visit . The Creative Commons Public Domain Dedication waiver (?

mmons.?org/?publi?cdoma?in/?zero/1.?0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Chan et al. BMC Cancer

(2021) 21:1338

of both hypertension and cancer [11, 12], and BP is

positively associated with white blood cell count [13].

Nevertheless, confounding by social and environmental factors could give rise to the observed associations

[14]. Mendelian randomization (MR), by using genetic

variants randomly allocated at conception as instrumental variables, is less susceptible to confounding

than conventional observational studies [15]. In this

MR study using two-sample methods, we assessed the

effects of systolic and diastolic BP on total cancer as

well as on 17 common site-specific cancers, by applying replicated genetic instruments for BP to large population-based cohorts. For validation, we included large

genetic consortia for breast, prostate and lung cancer.

We also used multivariable MR [16, 17] to mitigate

potential pleiotropic effects via obesity and smoking.

Methods

Genetic instruments for blood pressure

We extracted strong (P < ?

5x10-8), independent (r2 <

0.001) and externally replicated single nucleotide polymorphisms (SNPs) predicting BP from a meta-analysis

of genome-wide association studies (GWAS) for BP

traits totaling 757,601 participants of European ancestry (mean age 56.0 years, 54.7% women) [18], consisting

of 458,577 individuals from the UK Biobank excluding pregnant women (n=372) and individuals who had

withdrawn consent (n=36) [19], and 299,024 individuals from an enlarged dataset of the International Consortium for Blood Pressure (ICBP) with 77 cohorts [20].

Independent replication included 220,520 individuals from the Million Veteran Program [21] and 28,742

individuals from the Estonian Biobank of the Estonian

Genome Center University of Tartu [22]. Participants

on BP lowering medication had their BP values adjusted

by adding 15 and 10 mm Hg to systolic and diastolic

BP, respectively [23]. The UK Biobank analysis used a

linear mixed model [24], adjusted for age, ?age2, sex and

body mass index (BMI), with genomic control applied

at the study level to correct for inflation due to population stratification and cryptic relatedness [25], followed

by fixed-effect meta-analysis with the ICBP summary

statistics which also adjusted for the same covariates.

The pooled mean (standard deviation (SD)) systolic and

diastolic BP were 138.4 (20.1) and 82.8 (11.2) mm Hg,

respectively. The BP GWAS adjusted for BMI, which

could bias the estimates of genetic variants on BP if

genetic variants or environmental factors driving both

BMI and BP exist [26], and potentially the MR estimates

and/or instrument selection. We repeated the analysis

for total cancer using genetic predictors from the UK

Biobank, which did not adjust for BMI.

Page 2 of 9

Genetic associations with total and site?specific cancers

Genetic associations with total cancer (phenocode:195)

were obtained from a pan-ancestry GWAS of the UK

Biobank [27], with lifetime cancer occurrence ascertained

from linked medical records (hospital inpatient data and

death registry) including both prevalent and incident

cases [28]. MR studies evaluate lifelong effects of an exposure, and so necessitate the inclusion of lifelong cases

in consideration of potential selection bias [29]. Of the

441,331 participants included, genetic associations were

provided for the 420,531 (95.3%) individuals of European

ancestry to minimize confounding by population stratification. Non-cases were individuals without a diagnosis

of primary or secondary cancer, nor a history of radio- or

chemotherapy. The analyses used the Scalable and Accurate Implementation of Generalized mixed model, which

accounts for sample relatedness and extreme case-control

ratio [30], and adjusted for age, sex, age*sex, ?age2, ?age2*sex

and the first 10 principal components (PCs).

Genetic associations with site-specific cancers were

obtained from the largest available pan-cancer GWAS

[31], which provides summary genetic associations with

17 cancers for 475,312 individuals of European ancestry from the UK Biobank and the Kaiser Permanente

Genetic Epidemiology Research on Adult Health and

Aging (GERA) [32, 33]. Lifetime cancer occurrence was

ascertained from linked medical records with the latest

diagnosis in August 2015 in the UK Biobank and June

2016 in GERA, which were converted into the third revision of International Classification of Diseases for Oncology (ICD-O-3) codes and classified according to organ

site based on the U.S. National Cancer Institute Surveillance, Epidemiology, and End Results Program recode

paradigm [34]. The median age at diagnosis was lowest

for cervical cancer (37 and 38 years in the UK Biobank

and GERA, respectively) and highest for pancreatic cancer (66 and 76 years). Individuals with multiple diagnoses

were only recorded for their first cancer. Non-cases were

cancer-free individuals, i.e., those who did not have any

cancer diagnosis, self-reported history of cancer or cancer as a cause of death. For sex-specific cancer (breast,

cervix, endometrium, ovary, prostate and testis), samesex non-cases were used. Summary genetic associations

are available for bladder, breast, cervix, colon, esophagus/

stomach, kidney, lung, lymphocytic leukemia, melanoma,

non-Hodgkin lymphoma, oral cavity/pharyngeal, ovary,

pancreas, prostate, rectum and thyroid. The analyses

were conducted separately for each cohort using logistic

regression, adjusted for age, sex, the first 10 PCs, genotyping array (UK Biobank only) and reagent kit for genotyping (GERA only), followed by meta-analysis. Standard

error (SE) of the SNP-outcome association were estimated from the p-value [35], as it was not provided.

Chan et al. BMC Cancer

(2021) 21:1338

For validation of potentially small effects, we additionally included large genetic consortia of leading cancers

[36], i.e., breast (133384 cases and 113789 non-cases)

[37], prostate (79194 cases and 61112 non-cases) [38] and

lung (10246 cases and 38295 non-cases) [39], which have

larger number of cases and do not overlap with the UK

Biobank or GERA.

Estimates were aligned on the same effect allele for

BP and cancer. Effect allele frequency (EAF) was not

provided for pan-cancer, so we used the UK Biobank

EAF which constituted 86% of the participants. Palindromic SNPs with ambiguous EAF, i.e. >0.42 and

97% suggests minimal bias of the MR estimates by confounding of exposure on outcome in overlapping samples

[44], as here. The proportion of phenotypic variance (?r2)

explained by the genetic instruments was calculated as

?beta2*2*MAF*(1-MAF), where beta is the SNP-phenotype

association standardized to the phenotypic variance and

MAF is the minor allele frequency of the SNP [45]. Power

calculations were based on the approximation that the

sample size for an MR study is the sample size for exposure on outcome divided by the ?r2 for genetic instruments

on exposure [46], using an online tool [47].

We used the inverse-variance weighted (IVW) metaanalysis, with multiplicative random effects, which

assumes balanced pleiotropy [48], of the SNP-specific

Wald estimates, i.e., the SNP-outcome association

divided by the SNP-exposure association, as the main

analysis. We also conducted sensitivity analyses using the

weighted median [49] and MR-Egger [50]. The weighted

median assumes 50% of the weight is from valid SNPs.

MR-Egger is robust to genetically invalid instruments

given the instrument strength Independent of direct

effect (InSIDE), i.e., the instruments do not confound

Page 3 of 9

exposure on outcome, and the NOME assumption is

satisfied. A zero MR-Egger intercept indicates evidence

of lack of such genetic pleiotropy. Some of the genetic

instruments for BP was previously shown to be associated with confounders of BP and cancer, mostly for

anthropometrics and a few for lifestyle [18], so we used

multivariable MR to estimate the effects of BP on cancer

independent of BMI or ever-smoking using IVW or MREgger if the intercept was non-zero. We obtained genetic

associations with BMI and ever-smoking from Yengo

et al. [51], and the Social Science Genetic Association

Consortium [52], respectively. We dropped correlated

predictors of BP and BMI or ever-smoking. We estimated

the Sanderson-Windmeijer multivariate F-statistic and

modified Q statistic to assess conditional instrument

strength and heterogeneity, taking into account the phenotypic correlation [53], using estimates from the UK

Biobank [54].

All analyses were performed using R (version 4.0.1, The

R Foundation for Statistical Computing Platform, Vienna,

Austria). We used the R packages ¡°TwoSampleMR¡±,

¡°MedelianRandomization¡± and ¡°MVMR¡±. Given the number of cancer outcomes considered, a two-sided p-value

below the Bonferroni-corrected significance threshold

0.0014 (0.05/2 BP traits*18 cancer outcomes) was used.

Results

There were 272 and 267 strong, independent and replicated SNPs predicting systolic (Supplementary Table S1)

and diastolic (Supplementary Table S2) BP with a mean

(range) F-statistic of 83.2 (29.3 ¨C 612.4) and 90.7 (30.0 ¨C

818.1), and ?I2 of 92.5 and 93.8%, respectively (Table 1).

These SNPs explained approximately 2.59 and 2.96% of

the variance of systolic and diastolic BP, respectively. At

5% alpha, this study has 80% power to detect an odds

ratio (OR) of about 1.09 for total cancer, and from 1.13

for breast to 1.60 for thyroid cancer per SD of BP (Supplementary Table S3). We obtained 539 and 92 strong (P

< ?5x10-8) and independent (r2 < 0.001) SNPs predicting

BMI and ever-smoking, respectively. Both BMI (Supplementary Fig. S1) and ever-smoking (Supplementary

Fig. S2) were positively associated with total cancer. The

Sanderson-Windmeijer multivariate F-statistics were

at least 33.8 for systolic and 36.9 for diastolic BP when

adjusted for BMI, and at least 75.3 for systolic and 80.8

for diastolic BP when adjusted for ever-smoking.

Figure 1 shows the associations of systolic and diastolic

BP (per 1-SD increment) with total cancer. Overall, systolic (OR 0.98 [95% confidence interval (CI) 0.89, 1.07]

and diastolic BP (OR 1.00 [95% CI 0.92, 1.08]) were not

associated with total cancer. Repeating the analysis using

genetic instruments unadjusted for BMI, or sensitivity analysis using the weighted median, MR-Egger and

Chan et al. BMC Cancer

(2021) 21:1338

Page 4 of 9

Table 1 Genome-wide association studies of total and 17 site-specific cancers

Outcome

No. of SNPs used

(systolic/diastolic)

Total sample size

Total cancer

272/267

Bladder

270/267

Breast

No. of cases

No. of non-cases

UK Biobank

GERA

Total

UK Biobank

GERA

Total

405243

37387

-

-

367856

-

-

412592

1550

692

2242

359825

50525

410350

269/266

237537

13903

3978

17881

189855

29801

219656

Cervix

270/267

226219

5998

565

6563

189855

29801

219656

Colon

270/267

414143

2897

896

3793

359825

50525

410350

Endometrium

270/267

221693

1414

623

2037

189855

29801

219656

Esophagus/stomach

270/267

411441

929

162

1091

359825

50525

410350

Kidney

270/267

411688

1021

317

1338

359825

50525

410350

Lymphocytic Leukemia

270/267

411202

594

258

852

359825

50525

410350

Lung

270/267

412835

1728

757

2485

359825

50525

410350

Melanoma

270/267

417127

4271

2506

6777

359825

50525

410350

Non-Hodgkin¡¯s Lymphoma

270/267

412750

1760

640

2400

359825

50525

410350

Oral cavity/pharyngeal

270/267

411573

930

293

1223

359825

50525

410350

Ovary

270/267

220915

1006

253

1259

189855

29801

219656

Pancreas

270/266

411013

471

192

663

359825

50525

410350

Prostate

268/267

201486

7441

3351

10792

169970

20724

190694

Rectum

270/267

412441

1808

283

2091

359825

50525

410350

Thyroid

270/267

411112

527

235

762

359825

50525

410350

Abbreviations: GERA Kaiser Permanente Genetic Epidemiology Research on Adult Health and Aging, SNP single nucleotide polymorphism

multivariable MR gave similar results (Supplementary

Tables S4 and S5).

Systolic and diastolic BP were not significantly associated with any of the 17 site-specific cancers in the metaanalysis of the UK Biobank and GERA. Some associations

Systolic

MR method

IVW

IVW (genetic instruments unadjusted for BMI)

Weighted median

MR-Egger

Multivariable IVW (BMI-adjusted)

Multivariable IVW (smoking-adjusted)

at nominal significance were observed for kidney cancer

and melanoma (Fig. 2). Similarly, no significant associations were observed for breast, prostate or lung cancer (Fig. 3) in the consortia, although systolic BP was

nominally associated with lung cancer. Using other MR

Diastolic

OR (95% CI)

p-value

0.98 [0.89, 1.07]

0.619

1.00 [0.92, 1.08]

0.953

0.99 [0.92, 1.06]

0.755

1.01 [0.95, 1.08]

0.763

0.96 [0.85, 1.08]

0.472

1.01 [0.91, 1.12]

0.864

0.90 [0.72, 1.12]

0.339

0.89 [0.73, 1.08]

0.253

0.99 [0.90, 1.08]

0.755

1.02 [0.94, 1.11]

0.644

0.99 [0.90, 1.08]

0.773

1.00 [0.92, 1.09]

0.985

0.7

0.8 0.9 1.0 1.1 1.2 1.3

OR (95% CI) per SD increase of blood pressure

Fig. 1 Mendelian randomization (MR) estimates of systolic and diastolic BP on total cancer. BMI, body mass index; IVW, inverse variance weighting

Chan et al. BMC Cancer

(2021) 21:1338

Page 5 of 9

Systolic

Cancer site

Diastolic

OR (95% CI) p?value

Bladder

1.07 [0.79, 1.45] 0.658

1.05 [0.80, 1.38] 0.729

Breast

1.00 [0.88, 1.14] 0.976

1.03 [0.91, 1.15] 0.669

Cervix

0.88 [0.75, 1.04] 0.134

0.92 [0.79, 1.07] 0.279

Colon

0.92 [0.75, 1.13] 0.447

0.91 [0.75, 1.10] 0.334

Endometrium

0.91 [0.69, 1.20] 0.495

0.82 [0.64, 1.06] 0.128

Esophagus/stomach

1.17 [0.82, 1.68] 0.385

1.16 [0.83, 1.61] 0.394

Kidney

1.42 [0.99, 2.03] 0.059

1.29 [0.92, 1.79] 0.135

Leukemia

0.82 [0.55, 1.25] 0.362

1.02 [0.69, 1.52] 0.916

Lung

1.00 [0.77, 1.31] 0.973

0.92 [0.72, 1.17] 0.484

Melanoma

1.19 [1.01, 1.40] 0.036

1.27 [1.06, 1.51] 0.008

Non?Hodgkin's lymphoma

0.98 [0.77, 1.25] 0.885

0.77 [0.61, 0.98] 0.035

Oral cavity/pharyngeal

1.28 [0.90, 1.83] 0.173

1.32 [0.95, 1.83] 0.096

Ovary

0.89 [0.62, 1.27] 0.516

0.79 [0.57, 1.11] 0.172

Pancreas

1.12 [0.69, 1.81] 0.642

0.98 [0.61, 1.56] 0.923

Prostate

1.05 [0.90, 1.23] 0.544

1.04 [0.89, 1.21] 0.620

Rectum

0.84 [0.63, 1.12] 0.243

0.86 [0.65, 1.14] 0.296

Thyroid

0.97 [0.62, 1.52] 0.893

1.20 [0.77, 1.88] 0.416

0.5

1.0

1.5

2.0 2.5

OR (95% CI) per SD increase of blood pressure

Fig. 2 Inverse-variance weighted Mendelian randomization estimates of systolic and diastolic BP on 17 site-specific cancer

methods gave similar results for site-specific cancers

(Supplementary Tables S4-6). Systolic and diastolic BP

were not associated with asthma (Supplementary Fig. S3).

Discussion

Consistent with secondary analyses of RCTs [9], but less

consistent with observational studies [5¨C8], this MR

study found little evidence of BP increasing risk of cancer.

However, BP was nominally positively associated with

kidney cancer [55, 56], and possibly melanoma [57]. As

expected, BP was not associated with asthma.

This is the first MR study that has comprehensively

evaluated the effect of BP on cancer. The main strength

of the study is the MR design which minimizes confounding [58]. Long-term exposure to common risk factors

for cancer [14], such as socio-economic position and all

it entails, including smoking [59], alcohol consumption

[60], diets promoting obesity [61], and air pollution [62]

are known to elevate BP. So, previous observational findings showing higher BP positively associated with risk of

total and some site-specific cancers [5¨C8], might be due

to confounding by these factors. Sustained hypertension

leads to compensatory vascular hypertrophy involving

Angiotensin II mediated by various growth factors [63].

Angiotensin II receptors are found in high density in the

kidney responsible for BP regulation [64]. A previous MR

 ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download